Agent

An Agent is LanSphere’s LLM-centric app type: the model plans tasks autonomously from user intent, flexibly calling plugins, Workflows, Knowledge Bases, MCP Services, and more — coordinating resources to get the task done, with no fixed flow for you to pre-orchestrate.

This article covers the Agent’s positioning and use cases, each configuration on the “App Development” page, how to call a published Agent from a Workflow, logs and the dashboard, and publishing channels. It is for the Developer role and above.

Use Cases

Agents fit tasks with a clear goal but no fixed path, where the model must decide for itself. Typical scenarios:

• Personal office assistant: Check calendars, send messages, draft documents on request — choosing which capability to call on its own.
• Cross-system business handling: The user states a need in one sentence; the Agent breaks the task down and chains tools across systems to complete it.
• Calendar and message management: With the Lansenger MCP toolset, query and create events, and send private or group messages.

How It Differs from Chatflow

Both Chatflow and Agent serve multi-turn conversation, but the orchestration philosophy differs:

Tip: When the path is predictable, use a Chatflow — execution is more controllable. When the task is open-ended and needs flexible orchestration across capabilities, use an Agent.

Creating an Agent

1. Click “Agent” in the left menu (visible only to the Developer role and above).
2. Create a new Agent: app name (required, 1–30 characters; Chinese characters, upper/lowercase letters, digits, and hyphens only); description optional (up to 300 characters); custom avatar upload supported.
3. Confirm to enter the app details page.

The Agent module and the Workflow module keep data independent of each other. The app details page includes “App Development,” “Conversation Logs,” “Data Dashboard,” “Publishing Channels,” and more.

App Development

The “App Development” page provides the following configurations, explained one by one.

Model Selection and Configuration

Choose the LLM that serves as the Agent’s core and complete its parameter configuration as prompted on the page. Available models come from those connected and enabled in System Settings → “Models.” To add a provider or adjust default models, ask an Admin to configure it in “System Settings → Models.”

Prompt Writing

The prompt governs how the AI responds. It is for internal use only and invisible to end users. You can embed form variables in the prompt; they are replaced with what the end user fills in at run time.

Here is a sample prompt for an enterprise office assistant (use the variable syntax shown on the App Development page):

# Role
You are the organization's smart office assistant, serving its employees. You answer common administration, HR, and IT questions and help with day-to-day tasks.

# User info
- Name: {{Name}}
- Department: {{Department}}

# Response rules
1. Only answer questions related to the organization's business; for unrelated questions, politely explain your scope of service.
2. For policy and process questions, answer based on Knowledge Base content first and cite the source.
3. If you are not sure about something, say so plainly. Never make things up.
4. Address the user respectfully; keep a professional, concise tone; tailor answers to the user's department where relevant.

# Output format
- Give the conclusion directly for general questions, with bullet points where needed.
- For handling procedures, output numbered steps.

Form Configuration

The form collects info the end user fills in before the conversation starts (e.g., name, department, request type), producing form variables that automatically replace the corresponding variables in the prompt. The “form + prompt variables” combination lets the Agent grasp necessary background without asking.

Knowledge Base Association

Once a Knowledge Base is associated, the Agent can retrieve from it when answering, keeping answers grounded. For creating and maintaining Knowledge Bases, see “Knowledge Base.”

Tool Configuration

Attach callable tools to the Agent — three categories:

• Plugins: Call System Plugins and Custom Tools; for installation and creation, see “Plugins & MCP.”
• Workflow: Call Workflows published as components (from “Collaboration Components → Workflow”), packaging a fixed flow as one of the Agent’s capabilities.
• MCP Services: Call tools provided by services added under the “MCP Services” menu — e.g., messaging, calendar, and meeting capabilities in the Lansenger MCP toolset.

Once attached, the LLM decides which tool to call and when, based on user intent.

Skill Orchestration

In the “Skills” category of Tool Configuration, check installed skills to attach them to the Agent. A skill is a reusable capability unit — official skills are built in, and custom import and creation are supported. See “SKILL.”

Once attached, the LLM decides when to call them based on user intent. You can also allow or forbid specific skills per Agent to control its capability boundary.

File Configuration

Configure file-related capabilities as prompted on the page.

Error Handling

Configure how the Agent responds and handles runtime errors, as prompted on the page.

Calling an Agent from a Workflow

A published Agent can be reused by a Workflow as a node capability:

1. In the Workflow canvas’s node list, click the “Agent” node.
2. In the node configuration, single-select the Agent to attach from the published-Agent list.
3. Once wired, when the Workflow runs to this node, it calls the selected Agent to handle the work.

This fits “embed a stretch of autonomous decision-making inside a fixed flow” — e.g., let the Workflow prepare the data first, then hand it to the Agent for analysis and a conclusion.

Conversation Logs and Data Dashboard

• Conversation Logs: Run records organized by conversation, including message counts and, per message, input, output, elapsed time, Token usage, and user feedback. You can mark answers Like/Dislike and add annotated replies to keep improving answer quality.
• Data Dashboard: Visual charts of success rate, average response time, Token usage, tool-call distribution, and more, filterable by time range.

See “Logs & Monitoring” for details.

Publishing Channels

Agents can be published to the App Square, and web link / API access channels can be enabled in the “Publishing Channels” tab. The headline channel is publishing as a Lansenger AI Assistant: after publishing, the system automatically creates the app and bot instance on the Lansenger side and lists it in the Lansenger client’s “AI Assistant Marketplace,” ready for end users inside Lansenger.

Before publishing as a Lansenger AI Assistant:

• The publishing account must be linked to Lansenger; if not, follow the pop-up to the Lansenger authorization center — publishing runs automatically once QR-code authentication passes.
• Publishing requires the Admin role or above.
• The assistant title defaults to the app name (within 10 characters) and the description defaults to the app description (within 20 characters).
• End users must complete data authorization on first use and can only chat after approving it.

After a successful publish, the button changes to “Published” and cannot be repeated. For full channel flows and Version Management, see “Publishing Channels.”

Related Reading

• “Chatflow”: Multi-turn conversational apps with predictable flows.
• “Workflow”: Call a published Agent via the “Agent” node in a Workflow.
• “SKILL”: Installing, creating, and attaching skills to Agents.
• “Plugins & MCP”: Where plugins and MCP Services in Tool Configuration come from.
• “Publishing Channels”: Full flows for publishing as a Lansenger AI Assistant and other channels.